[tt] [wta-talk] Kurzweil critique
Eugen Leitl
<eugen at leitl.org> on
Mon Nov 5 21:30:18 UTC 2007
----- Forwarded message from Michael Anissimov <michaelanissimov at gmail.com> -----
From: Michael Anissimov <michaelanissimov at gmail.com>
Date: Mon, 5 Nov 2007 13:19:26 -0800
To: sl4 at sl4.org, Extropians <extropy-chat at lists.extropy.org>,
World Transhumanist Association Discussion List <wta-talk at transhumanism.org>
Subject: [wta-talk] Kurzweil critique
Reply-To: World Transhumanist Association Discussion List <wta-talk at transhumanism.org>
On my blog, someone anonymous posted a long rebuttal
([1]http://www.acceleratingfuture.com/michael/blog/?p=606#comment-1022
90) of Kurzweil's ideas that I think some people may be interested
in. Note that just because I post it doesn't mean I agree or disagree
with any specific claims therein. Enjoy!
~~~
Ray Kurzweil is an obvious crackpot. He's nothing but a much
better-educated version of your typical ufologist.
The claims he makes about current technology are provably false, so we
shouldn't be surprised that the claims Kurzweil makes about future
technology qualify as delusional.
There exist so many clear-cut examples of Kurzweil's claims being
obviously and flagrant false that it's hard to choose just a few, but
one good example is Kurzweil's flagrantly false assertion that "We
understand the human ear and we have reverse engineered it," referring
to cochlear implants. This is not just wrong, it's widely known to be
wrong.
Roughly 1/3 of cochlear implants work well enough for the recipients
to understand speech in cases where there isn't overlapping
conversations or ambient noise. However, even in those best-case
scenarios, the cochlear implant never works 100% of the time, and
basically functions as an aid to lip-reading. So even in the most
successful cases, people with high-functioning cochlear implants need
to lip-read some of the time to understand human speech. In another
1/3 of the cases cochlear implants work at a low-functioning level and
it's possible to understand some speech, but music and other sounds
don't come through well. (in the best high-functioning cases, cochlear
implants not only allow the recipient to hear music, but to enjoy it.)
And in 1/3 of the cases cochear implants don't work at all.
[2]www.johnhorgan.org/work16.htm
It should be emphasized that scientists do not understand why cochlear
implants work well in some recipients and don't work at all in others.
It's not the technology since the implants are identical.
It should also be pointed out that whenever anyone gets a cochlear
implant, they initially go through a long period of several months in
which they perceive nothing but noise coming from the implant. The
brain gradually adjusts to the signals and eventually deciphers them
(in cases where the implant works) and over a long period of time, in
the best cases, recipients can hear not only pitch but also timbral
differences. However, cochlear implant recipients who lost their
hearing as children or as adults report that even in the best case,
cochlear implants produce input that sounds nothing like ordinary
hearing.
So Kurweil's claim that scientists have mapped the brain and
understand how much of it functions are provably false. Scientists
have not mapped the brain even partially. We still don't know all the
functions of (for example) the left temporal lobe. Scientists do NOT
understand how even the simplest parts of the brain, like the auditory
cortex, function at least, not well enough to reverse-engineer them.
As for nanotechnology and hard AI, those fields have run into brick
walls so complete that there's no more point in discussing those
delusions than in debating the claims of scientologists or alchemists.
[3]http://www.skeptic.com/the_magazine/featured_artic
les/v12n02_AI_gone_awry.html
Science is making progress and technology has produced many advances.
But the kinds of advances Kurzweil is talking about are not just
futuristic, they belong to the realm of hallucinogenic self-delusion
like GM's nuclear powered Nucleon concept car, a robot with
human-level intelligence and manual dexterity in every home, a
personal helicopter for everyone, personal jet packs, flying cars, and
other seemingly drug-induced fantasies out of the TV kiddies' cartoon
The Jetsons.
Kurzweil's claims about enhancing intelligence through genetic
engineering in particular show his desperate ignorance of basic
molecular biology and population statistics and cognitive psychology.
No one knows what intelligence is or how to measure it and the
evidence for that failure is overwhelming:
For most of the 20th century, intelligence was viewed as an
all-purpose, monolithic power, christened g by psychologist Charles
Spearman. Creativity was believed to be a side effect of a high level
of general intelligence - a mark of big g. The father of the
standardised-testing industry, Lewis Terman, created the
Stanford-Binet Intelligence Scale to quantify this power. He launched
the longest scientific study in history, Genetic Studies of Genius, to
track the accomplishments of highly gifted grade-school children
through the course of their lives. His hope that an impressive IQ
score would augur groundbreaking accomplishments in science and art,
however, didn't pan out. His young Termites, as he affectionately
called them, did end up earning slots at better universities and
getting hired for executive positions, often with help from Terman.
They gave the world two memorable inventions: the K ration and I Love
Lucy. (Both Ancel Keys, who perfected single-meal pouches for the US
Army, and Jess Oppenheimer, the creator of the popular TV show, were
Termites.)
For the most part, however, real genius slipped through Terman's net.
None of his prodigies won major scientific prizes or became important
artists, while two students excluded from the study for having
insufficient test scores, William Shockley and Luis Alvarez, went on
to earn Nobels.
[4]http://www.flatrock.org.nz/topics/education/a_rage _to_master.htm
A test that allegedly measures "intelligence" but sieves out two
future Nobel prize winners in the sciences constitutes such an
obviously grotesque failure that hardly anything else need be said on
the subject of testing for intelligence.
Clearly, we can't reliably test intelligence. We don't have a ghost of
a clue what intelligence is, and we have no idea how to about figuring
out how to determine what intgelligence is.
What we do know about measured IQ is that it is not correlated with
achievement or general problem-solving ability. Marilyn vos Savant,
the person with the highest recorded IQ, used to author a puzzle
column for a newspaper, and now works as an accountant for her
husband's business. Hardly a stellar record of achievement. You might
expect the highest IQ segment of the population to correlate with the
admissions to the top 50 colleges or the list of Nobel Prize winners
you'd be wrong. Dead wrong. Completely 100% wrong. Turns out a
Bulgarian woman with one ofhte highest recorded IQs can't even get a
job, much less admission into a top-50 U.S. university.
Nobel Prize winners tend to come from small colleges, not out of the
top 50 most prestigious colleges. Nobel Prize winners tend not to come
from the top half percent of the IQ test scorers Richard Feynman had
measured IQ of 120, much much lower than Marilyn Vos Savant or most of
hte pople in MENSA. The delusion that we know what intelligence is,
and therefore we can build smarter computers, and that those computers
will therefore be able to build even smtarter computers, is a chain of
errors as foolish and as crazy as the chain of errors involved in
claiming that lightning bugs are produced when lightning strikes a
bug.
We don't know what intelligence is. Even if we did know, there's no
evidence we can enhance it or replicate it. (We know perfectly well
what imgaination is can we enhance ir or replicate it?) Even if we
could enhance or replicate intelligence in silicon, there's no
evidence at all that a smarter-than-human computer would be able to
build a computer smarter than itself (and there's a huge mountain of
evidence showing that it couldn'tjust look how impossible it has been
for the smartest humans to produce computers smarter than themselves).
And even if superhumanly smart computers could produce computers
smarter than themselves, what's the evidence that they wouldn't just
sit around contemplaing beautiful paintings instead of interacting
with humans? Do really smart human spend their time explaining
themselves to ants? Why would superhumanly smart computers even bother
to interact with us, assuming they were possible whiich is unliikley
to the point of practical impossibility?
Nobel prize winners, asked about what produced their breakthroughs, do
not cite intelligence instead, they refer to qualities like
"imagination" and "persistence." Neither Ray Kurzweil nor any
molecular geneticist has suggested or shown any method of genetically
eningeering reliable enhancements to human creativity or persistence.
No one even has any idea how to measure these qualities
quantitatively, much less genetically enhance them, or even if they
can be genetically enhanced.
The usual kooks and cranks and flakes will of course erupt with
red-faced flatulent fury to shriek "that article from Skeptic magazine
you linked to doesn't prove anything!"
That's a lie.
Moreover, it's simple and easy to prove that it's a lie.
The article proves that none of the myriad claims made by AI
researchers have ever panned out, it proves that every single one of
the most prestigious current AI researchers with tenured positions as
head of the best cutting-edge AI resarch labs in the finest
universities in the world all believe "AI is brain-dead" and "AI has
hit a brick wall."
The article from the Skeptic magazine cited above proves that there
are not just one, but many incredibly hard problems facing AI research
problems so unsolvable, so refractory, so shockingly intransigent,
that no one has even been able to suggest even a hypothetical way to
get around them, much less make progress in AI and genetic engineering
of human intelligence or build Drexlerian nanotech assemblers, by
solving them. These problems include the frame problem and the
combinatorial explosion search problem for expert systems and the
self-reference problem for AI, the problem of junk DNA and the RNA
world paradigm and the really really tough problem of
reverse-engineering emergent systems for genetic engineering, and the
problem of molecular stiction and Brownian bombardment and the
destruction by Brownian forces and Van Der Waals forces and molecular
folding of the paper-tape-type ecnoded instructions required for a
rod-logic atomic level computer to work and be programmable in a
general Von Neuman sense.
Before the kooks and cranks and flakes who deny that Kurzweil is
spouting gibberish continue with their rants, they need to do the
following:
[1] Show us a working AI computer program which solves the frame
problem. Not just a diagram, not just pseudo-code, not just a research
paper on how to write such a program a working AI program that solves
the frame problem. Show us such an example, or shut up because you're
an ignorant liar.
[2] Show us a working automated translation program that reliably
takes in natural language and reliably spits back out idiomatic
English without grammatical or semantic errors. Not just a program
that works on 50% of the words in sentences, not something that needs
huge amounts of human intervention to work, not pseudo-code, not a
white paper on how to write such a program, but an actual working AI
program. Show us that, or shut up because you're an ignorant liar.
[3] An AI program that reads a novel and summarizes the book in a book
report that's accurate and succinct. Not just pseudo-code, not just a
research proposal, but an actual working program. Show us that, or
shut up, because you're an ignorant liar.
[4] A computer program that can listen to a piece of music and tell us
whether it's any good. In other words, a computer program that can
realiably tell the difference between randomly-generated junk and a
pop tune. Any human can tell the difference in 3 seconds, but no
computer can. Once again, don't just provide pseudo-ccde, not just a
research proposal, but an actual working program. Show us that, or
shut up because you're an ignorant liar.
[5] An AI program which can negotiate a labor agreement. Not just
pseudo-code, not a proposal, but an actual working program. Show us
that, or shut up because you're an ignorant liar.
Every single time the kooks and cranks and flakes who deny that
Kurzweil is a crackpot get asked to show any of these actual working
computer programs, they always give evasions and excuses. They
backpedal and fum-fuh and spin long-winded elaborate incoherent
stories to explain why they can't give us any evidence.
In short, Kurzweil and his supporters when asked for evidence give the
same kind of response you get from ufologists or Bigfoot enthusiasts
or hollow earth proponents when you ask 'em for hard evidence of their
claims. They give you nothing nothing but smoke and mirrors, lies and
bullshit, incoherent excuses and vague assertions like "it may take
many years to produce results" or "we're just starting to reesarch
these areas." The exact same kinds of vague hand-waving you get when
you confront ufologists and ask them for proof of their wild claims.
As for the kooks and cranks and flakes who will claim "it's easy
enough to debunk all these claims that AI and genetic engineering to
enahnce human intelligence and nanotechnology don't work and aren't
working and never will work, but I don't have the time" you're lying
and I can prove it.
If you can debunk the assertion that these technologies don't work and
haven't worked and can't work, greatdo it. Do it now. Do it right now.
Give us the hard evidence that hard AI works. Give us the hard
evidence that nanotechnology works and produced operating Drexlerian
assembler. Give us the hard evidence that genetic engineering can
reliably enhance human intelligence. Give us that hard evidence that
claims about people "uploading their minds into computers" are
anything more than a foolishly ignorant delusion based on the fantasy
that Descartes' mind-body divide is actually real and that there
exists some magical intangible Platonic essence called "the mind"
that's distinct from and separable from the human body (meaning the
human brain).
Antonio Damasio, in his book Decartes' Error, has deep-sixed most of
the ignorant misconceptions on which hard AI is based. I.e., that
there exists some magical fluid called "mind" separate from the brain;
that human thought is primarily logical and rational rather than
emotion-based and arising from bodily states; that humans use logic to
solve problems, rather than intuition and experience; that thought
involves sequences of computations, rather than emotions; that the
brain is a mere piece of hardware for a pseudo-computer-program called
"the mind." Kurzweil and his followers seem to be aware of none of
this. They never mention Damasio's somatic-sensory hypothesis:
"Although I cannot tell for certain what sparked my interest in the
neural underpinnings of reason, I do know when I became convinced that
the traditional views on the nature of rationality could not be
correct. I had been advised early in life that sounds decisions came
from a cool head I had grown up accustomed to thinking that the
mechanisms of reason existed in a separate province of the mind, where
emotion should not be allowed to intrude, and when I thought of the
brain behind that mind, I envisioned separate neural systems for
reason and emotion But now I had before my eyes the coolest, least
emotional, intelligent human being one might imagine, and yet his
practical reason was so impaired that it produced, in the wanderings
of daily life, a succession of mistakes, a perpetual violcation of
what would be considered socially appropriate and personally
advantageous.
I began writing this book to propose that reason may not be as pure as
most of us think it is or wish it were, that emotion and feelings may
not be intruders in the bastion of reason at all: they may be enmeshed
in its networks, for worse and for better.
I wrote this book as my side of a conversation with a curious,
intelligent, and wise imaginary friend, who knew little about
neuroscience but much about life My friend was to learn about the
brain and about those mysterious things mental, and I was to gain
insights as I struggled to explain my idea of what body, brain, and
mind are about."
[5]http://serendip.brynmawr.edu/bb/damasio/descartes. html
Kurzweil and his followers never discuss the frame problem in AI when
they blithely rhadsodize about superhumanly smart silicon
intelligences:
[6]www.iscid.org/encyclopedia/Frame_Problem
Kurzweil and his sycophants never discuss the death of a patient in a
recent and relatively mild gene therapy attempt when they talk
blithely about genetically engineering much larger wholesale
transformations of human beings into superhumans:
[7]www.fda.gov/fdac/features/2000/500_gene.html
Ray Kurzweil and his toadies just ignore whole bodies of knowledge in
order to further their crackpot claims.
Show us the hard evidence for Kurzweil's extroarindary claims or shut
up.
Point us to a list of peer-reviewed articles in scholarly journals
providing hard experimental evidence that these technologies do work.
Everything else is bullshit.
Put up or shut up. Provide hard evidence that the hypothetical
technologies touted by Kurzweil actually could do what he claims they
could, or stand revealed as an ignorant crackpot and compulsive liar.
"Proof" means a peer-reviewed journal article by a reputable
scientists reporting verified and repeated experimental results.
Everything else is not proof.
I'm not interested in anecdotes, or just-so stories, or logical
arguments, or elaborate what-if scenarios those are the realms in
which scientologists and ufologists and other crackpots prefer to
operate.
Out here in the real world, we require proof before we believe a
claimand the more extraordinary the claim, the more exotraordinary the
amount and quality of the evidence required for us to believe it.
Ray Kurzweil has made not just one, but many, extraordinary claims. He
claims not just that hard AI will produce human-level intelligence,
but that it'll happen soon, and go on from there to produce superhuman
levels of intelligence. Ray Kurweil claims not just that we'll be able
to simulate the human mind in silicon, but that we'll be able to
upload our minds into computers, and that it'll happen soon. Ray
Kurzweil claims not just that we'll be able to reliably genetically
engineer traits like human intelligence which all the evidence shows,
if they're heritable at all, must be polygenic and emergent, but that
we'll be able to do it soon, and to reliably produce enhanced human
capabilities that go far beyond the human, and that this genetic
engineering will not have dire side effects like, oh, say, terminal
leukemia, or autism, etc.
Ray Kurzweil claims not just that we'll be able to overcome molecular
stiction and Brownian motion and the bombardment of phonons at the
atomic-level to produce working rod-logic molecular computers, but
that we'll be able to produce molecular assemblers capable of being
reliably programmed and that can tear apart any type of matter and
rebuild it into anything we like, and that this will happen soon.
This is tantamount not just to claiming that an evil Alien Xenu is
responsible for invivible thetans that cause all mental illnessbut
that Xenu is real and the earth is flat and there's a an alchemical
secret to turning lead into gold that anyone can use (and that doesn't
involve a cyclotron) and and the earth is hollow and full of Nazis
waiting to re-emerge and start WW III and that lizard men from Zeta
Reticuli use secret underground entrances to get into the White House,
where they plot to convert us all to Rosicrucianism.
Sane people demand hard evidence.
And when you get the truly wild claims of the kind of Ray Kurzweil has
made, we demand not just hard evidnece, but a veritable mountain of
bulletproof evidence before we'll believe claims this outlandish.
Yet what has Ray Kurzweil and his transhumanist extropian
Singularitarian followers given us?
Nothing. No hard evidence at all. Just a bunch of PR. Eric Drexler has
produced zero scientific research to support his claims, he's just
given a bunch of speeches and written some books. Hans Moravec has
produced no scientific research showing that his "bush robots" are
possible he's just written some books and given some interviews.
Folks, people who only write books and give interviews about fabulous
future developments aren't scientists, they're called "science fiction
authors." Science fiction is not reality. Don't confuse the two.
Have transhumanist extropians like Kurzweil and Moravec and Charles
Stross and Cory Doctorow given us even the level of hard evidence in
support of their claims that we would demand to convict a single
person of murder in court?
Nope. They haven't even given us that. Not even that much evidence.
To convict someone of murder in court, we demand forensic evidence and
eyewitness testimony, not just tall tales and might-be stories and
wild guesses. How much hard forensic evidence have we seen that hard
AI will fulfill its many promises?
Zero.
How much eyewitness testimony have we heard for working Drexlerian
assemblers and mind uploading and genetic engineering that produces
superhumanly smart people?
None.
So we haven't even gotten a minimal level of hard evidence, comparable
to what you'd demand to convict someone in court of murder, out of Ray
Kurzweil and his Singularitarians, in support of their outlandish
end-of-the-world-as-we-know-it Singularity predictions.
Yet any sensible person would demand far more and far better evidence
than you'd demand just to convinct someone of murder, in order to get
us to believe their extraordinary transhumanist claims and uploading
minds and creating supermen from DNA tweaks.
After all, people commit murder every day. Murder is commonplace
compared to mind uploading. Murder is quotidian - compared to creating
a superhumanly smart computer. We see murders all the time, we read
about them daily, we hear about them on the news. No one has ever seen
a superhumanly smart computer. No one has ever shown a person
uploading his mind into a computer. No one has ever genetically
engineered a superhumanly smart human being. And yet Ray Kurzweil and
his transhumanist Singularitarians expect us to believe their much
more fantastic claims with much LESS evidence than a sensible rational
person would demand to convict a defendant in court of the far more
ordinary and vastly more credible crime of murder.
Does any of this ring a bell? Does anyone smell a rat? Doesn't anyone
see the scam that's going on here?
I want hard evidence for transhumanism and the alleged Singularity not
baseless assertions.
I want to see working computer programsnot just-so stories.
I want to see actual functioning robots that don't bang into walls and
that can recognize the difference between a dog and a volleyballnot
just wild claims.
I want to see a functioning AI program that does real-language
translation without appallingly stupid and shockingly obvious errors,
like turning the motto "Out of sight, out of mind," into "Blind and
insane," or mistranslating "The spirit is willing but the flesh is
weak" into "The liquor is good but the meat is rotten."
I want to see a working genetic engineering vector that reliably makes
a rat 200% smarter not just the sequencing of the rat genome.
I want to see a working Drexlerian assembler that can rip apart a
spoon and turn it into a miniature Sterling engine. Show it to me. Let
me see it working.
There are no such AI programs or robots or genetic engineering vectors
or nanotech assemblers..
There is no such hard evidence for Kurzweil's wild claims.
After 50-plus years of sustained effort by the smartest people on
earth, there has been ZERO progress in these areas. In the article
"There's Plenty Of Room at the Bottom," in 1959, physicist Richard
Feynman largely anticipated K. Eric Drexler's ideas from his 1987
Engines Of Creation. In the 50 years since Feynman gave his lecture,
we've seen zero progress in creating anything like what Feynman talked
about. No molecular machines capable of tearing apart molecules and
rebuilding 'em to spec. No Drexlerian assemblers. No programmable
virus-sized machines. No atomic-scale rod-logic computers. None.
Zilch. Zip. Diddly. Bupkiss. Nada. Zippo. Nothing.
Claim I'm stupid or lying?
Great. Show us the proof.
Let us see the hard evidence. Put up or shut up.
Pay attention, folks. Notice the scam here. Every single objection to
skeptical requests for evidence of transhumanist Singularitarian
predictions gets met with the exact same type of reasoning used by
ufologists and scientologists and Bigfoot fancier.
Ufologists claim not enough research has been done on UFOS and that's
why there's no evidence for alien abductions Ray Kurzweil claims not
enough research has been done on AI and nanotech and genetic
engineering, and that's why there's no hard evidence for superhumanly
smart computers and genetically-engineered supermen and mind uploading
and Drexlerian nanomachines that can rip apart matter at the atomic
level and rebuild it atom by atom. Exact same type of reasoning as
ufologists.
Bigfoot enthusiasts claim it hasn't been long enough to give us
evidence of Bigfoot's existence, but that we'll see lots of evidence
real soon now. Ray Kurzweil makes the exact same claim "it's early
days yet in AI research, we haven't been at it long enough to give us
proof of the inevitable triumph of superhuman hard AI"the exact same
argument as the Bigfoot crackpots.
Scientologists claim people who don't see dramatic cures for their
mental problems need to spend more money Ray Kurzweil and the AI and
nanotch crackpots also say that we haven't seen dramatic new results
like superhumanly smart computers and mind uploading because we need
to spend more money. And, just like the Scientologists, no matter how
much money we spend on AI and nanotech, it's never enough. We always
need to spend more money. More and more and more money, and never any
results. And what's the answer to any skeptic who objects? "You need
to spend more money." Just like Scientology.
Psychic "researchers" can never provide us with a definitive point at
which a sensible person can conclude "ESP is bullshit." No, they tell
us we just have to keep spending money on their fruitless experiments
that never produce results, we just have to keep supporting their
failed psychic research forever and ever, amen. Same thing with Ray
Kurzweil and his crew they can never provide us with a single
experiment, which, if it fails, means hard AI is dead. They can never
give us a single condition under which we could conclude that
Drexlerian nanotech is a degenerating research program and must be
abandoned. No, just like the psychic crackpots, Ray Kurweil and his
crew continually demand more and more money for their failed AI
efforts, more and dead-end research with no results, forever and ever,
and no matter how unbroken the string of failures, they can never
accept any evidence as being sufficient to disprove their claims ofr
superhumanly smart computers and genetically engineered supermen and
mind uploading.
After 50 years of concentrated effort by the greatest geniuses on
earth, the best AI programs today still get fed a sentence like "The
astronomer married a really hot star" and STILL can't answer "What
does the word `hot' mean in that sentence?"
The finest AI programs today get fed a sentence like "Mary saw a puppy
in the window and wanted it," and they still can't answer the simple
question: "Which one did Mary want the puppy, or the window?"
If you believe Kurzweil's bullshit and you've swallowed the
Singularitarian Kool-Aid, great show us computer programs that can
correctly answer the above questions.
Otherwise, shut up, because you're spouting ignorant tripe.
--
Michael Anissimov
Lifeboat Foundation [8]http://lifeboat.com
[9]http://acceleratingfuture.com
References
1. http://www.acceleratingfuture.com/michael/blog/?p=606#comment-102290
2. http://www.johnhorgan.org/work16.htm
3. http://www.skeptic.com/the_magazine/featured_artic
4. http://www.flatrock.org.nz/topics/education/a_rage
5. http://serendip.brynmawr.edu/bb/damasio/descartes.
6. http://www.iscid.org/encyclopedia/Frame_Problem
7. http://www.fda.gov/fdac/features/2000/500_gene.html
8. http://lifeboat.com/
9. http://acceleratingfuture.com/
_______________________________________________
wta-talk mailing list
wta-talk at transhumanism.org
http://www.transhumanism.org/mailman/listinfo/wta-talk
----- End forwarded message -----
--
Eugen* Leitl <a href="http://leitl.org">leitl</a> http://leitl.org
______________________________________________________________
ICBM: 48.07100, 11.36820 http://www.ativel.com http://postbiota.org
8B29F6BE: 099D 78BA 2FD3 B014 B08A 7779 75B0 2443 8B29 F6BE
More information about the tt
mailing list